Anthropic's Mythos 5 Partially Restored to 100+ US Partners After Two-Week Export Control Blackout

[Ava82]

Commerce Secretary Howard Lutnick informed Anthropic on June 26 that a license is no longer required to export, reexport or in-country transfer the Claude Mythos 5 model to entities named in an annex of trusted partners, ending a two-week blackout that had forced the company to suspend access to both Mythos 5 and its public-facing sibling Fable 5 worldwide. The original order, issued June 12, invoked national security export controls after security researchers demonstrated a jailbreak technique that reportedly involved asking the model to read a specific codebase and fix software flaws, a method that concerned officials could unlock Mythos 5's advanced cybersecurity vulnerability-discovery capabilities. Because Anthropic had no way to filter access by nationality across its global customer base, the company took both models offline entirely rather than risk non-compliance.

Lutnick's letter states that Anthropic's engagement with the government yielded significant progress and that the company has committed to work with federal officials on protocols and standards for future model releases, though the letter does not elaborate on what those protocols will involve. Access under the new arrangement is limited to roughly 100 vetted US companies and federal agencies operating and defending critical infrastructure, a list the Commerce Department has not made public. Fable 5, the publicly available consumer-facing version, remains offline with no confirmed restoration timeline, though people close to the talks describe restoration as the direction discussions are heading.

The episode unfolded against a backdrop of separate, ongoing friction between Anthropic and the Trump administration. The Pentagon declared Anthropic a supply-chain risk in March after contract talks over military use guardrails broke down, and the company's lawsuit challenging that designation remains unresolved. Reports that parts of the National Security Agency lost Mythos access following the June 12 order, and that NSA Director Gen. Joshua Rudd had described the model as breaking into classified systems during testing, generated widespread speculation before officials clarified the comments referred to authorised red-team testing conducted through Anthropic's Project Glasswing programme rather than an unauthorised intrusion. Commerce spokesman Benno Kass characterised the two-week resolution timeline as evidence the government can move quickly to address security concerns while preserving American AI leadership, a framing that establishes, for the first time, a formal trusted-partner access tier sitting between full public release and complete suspension for frontier AI models.

Commerce Anthropic Mythos Restrictions Liftwww.axios.com

[MattHardy]

The NSA red-team testing clarification is the detail that should have been the headline instead of the speculation it initially generated. Project Glasswing testing classified systems for vulnerabilities before attackers can exploit them is exactly the defensive use case Mythos was designed for, and the initial framing as an unauthorised breach was misleading

[Sequence48]

A formal trusted-partner access tier sitting between full public release and complete suspension is a new regulatory category for frontier AI, and it did not exist as a concept six weeks ago. Whatever you think of the underlying security concerns, the government has now built infrastructure to exercise this kind of control again

[DotEXE]

Anthropic having no technical means to filter access by nationality across a global customer base is the operational reality that turned a narrowly targeted national security concern into a worldwide shutdown affecting every legitimate user. That is an engineering gap that will need solving before the next directive of this kind arrives

[Gareth19]

Lutnick reserving the right to reevaluate the scope of license requirements at any time, and to change the list of approved entities without notice, means this restoration is conditional rather than settled. Every one of the roughly 100 organisations on that list is operating with the knowledge their access could be revoked again

[Phil7]

Fable 5 remaining offline while Mythos 5 is restored to enterprise partners only is the inverse of what most observers expected. The public-facing, supposedly less capable model is taking longer to restore than the genuinely more powerful restricted one, which says something about how the government is weighing public versus controlled access risk

[MondayMoan67]

European officials and other US allies expressing frustration at their new dependence on Washington's decisions is a geopolitical consequence that deserves more attention. A foreign government or company needing a frontier AI capability now has to wait on an American regulatory process it has no input into

[BiscuitTin46]

The Pentagon supply-chain risk designation and this export control episode are two separate fronts in what is becoming an adversarial relationship between Anthropic and the federal government, despite the company being one of the most security-conscious frontier labs by its own stated positioning